Deep Learning-Based Prediction and Classification of Microplastics in Water Samples

Microplastics have become a major environmental and public health issue due to the large amounts of polyethylene (PA) particles in the drinking water supply. This project will develop an automated detection and classification method for polyethylene microplastics in drinking water using deep learning. A data set provided by the California State Water Resources Control Board contains a combined set of high-resolution microscopic images and structured metadata about particles (e.g., particle size, color attributes, and geospatial sampling coordinates). A convolutional neural network (CNN) model will be developed using hierarchical feature extraction to classify PE microplastics compared to traditional machine learning approaches such as random forests (RF) and support vector machines (SVM). The results of the experimental evaluation demonstrate that the CNN model outperformed both traditional machine learning approaches with an accuracy (92%), precision (93%), recall (91%), F1-score (0.92) and area under the curve (0.95). Geospatial analysis results identify contamination hotspots which can be used by water managers for environmental monitoring and industry decisions. The proposed automated method provides a scalable, fast, low-cost means to detect microplastics on a large scale and provide real-time assessments of drinking water quality.